speed up relmatLmer?

[GuoanQi1996]

Hi, @variani Fitting a linear mixed model with kinship as random effect is a quite time-consuming task. I am trying to apply the relmatLmer on a dataset with about 500 samples and 30K "variants" (not the numeric variable but factorial), by fitting Model1 = relmatLmer(PHE ~ PC1+PC2+PC3+PC4+PC5+(1|ID), data, relmat = list(ID = Kin),REML=F,control = lmerControl(optimizer = "nloptwrap",optCtrl=list(xtol_abs=1e-6, ftol_abs=1e-6),calc.derivs = F)) and turn into a loop Model2 = update(Model1, .~.+(1|G)) which G is a single factorial variant. Then two models were compared by anova to get the significance of G. This takes about 20 hours, even i paralleled the loop by foreach in package doSNOW with 40 threads.

I have already followed some methods on the stack overflow to improve the efficiency, but the time used is still not ideal. Is there any way to speed up the regression?

[variani]

Hi,

Do you have any suggestions on how to improve lme4qtl for your type of variance-component test? Any reference would be helpful.

Your Model2 doesn't account for mean differences among levels of G. Is it a reasonable assumption?

Best, Andrey